Staying Long Electrons — and the Things That Produce Them

key points

The Weekender is my bi-weekly take on macro shifts and emerging themes. It’s not investment advice — or even our firm’s official view. I aim simply to inform, challenge, and maybe entertain. If you’d like this in your inbox every other Saturday morning via Northern Trust, subscribe to The Weekender.

Staying Long Electrons — and the Things That Produce Them

Sam Altman, CEO of OpenAI, reportedly wants the company to reach 250GW of compute capacity by 2033. That’s equivalent to 250 nuclear reactors — 2.5 times the number currently operating in the U.S. today. It’s roughly one-third of peak U.S. power consumption, or close to the total energy produced in India. It would also require 25 times the annual supply of Nvidia AI chips (if I’m counting correctly). Not bad — for one company. I won’t attempt to critique this. My mind works linearly — if it works at all — and tends to shut down around the fourth turn of the exponential function.

So I look for signal in the actions of others. Like the deep-pocketed, price-inelastic U.S. government, which recently backed Lithium Americas (up 120% last week).

A reminder: success in the ethereal realm demands extraordinary support — and supply — from the physical one. A world still priced on its past. Not its present. And certainly not its future.

The Race for Electrons

If AI is a race for electrons, China may be in the lead.

“China’s electricity capacity exceeds that of the U.S., the EU, and India combined.” — Mining Weekly

Bloomberg New Energy Finance (BNEF) projects $15.8 trillion in global grid investments ahead. That implies demand for mountains of uranium, natural gas, copper, aluminium, silver, lithium, and steel. Supply is already constrained (see Grasberg Outage), and now increasingly weaponized in a multi-polar world. Add a new inelastic, politically driven buyer to the mix and you have the potential for a perfect storm. The sector driving this demand currently trades at roughly seven times forward sales, compared to just 1.3 times for the S&P Metals and Mining Select Index. That kind of divergence raises a question: Is mean reversion still a thing?

Uncomfortably Long

The talk in London this week: a Financial Times article on legendary tech investor James Anderson turning cautious on AI and Nvidia. Cautious, yes — but still a holder. Like anyone who lived through the dotcom bust, he’s likely uneasy watching familiar patterns play out — eye-watering valuations for OpenAI, Anthropic, and vendor/circular financing channelled through Nvidia. Echoes of 2000, minus the nostalgia.

But there are also differences: Nvidia’s circular financing is backed by cash flow, not debt, and met by demand that reportedly outstrips capacity (25 times, if you believe Mr. Altman!). So zoom out, broaden the aperture, and search for counterfactuals — like Anderson’s other interview, in the same newspaper, on the same topic and company, suggesting Nvidia could reach a $50 trillion market cap within a decade. That’s a lot of zeroes. My head hurts. I’ve stopped counting.